Hacker News Reader

Comment by FredPret | Hacker News Reader

They have working Linux drivers. I tried setting up Tensorflow with an AMD card once. I send a thousand dollars to Nvidia after a week

View on HN

They might have them, but they're closed source and when there are issues, you're completely fucked.

As opposed to the everyday case with AMD, where you have adventures all the time?

They both need to get this 1000% better stat. I think it's probably the best ROI the CEO of AMD can get - put a team of engineers on Linux drivers full time, increase your AI market share... and triple your market cap.

I've had no problems with Nvidia Linux drivers themselves in the last 15 years or so. Yes, they don't play well with the rest of the ecosystem. Using Optimus laptops was a nightmare but you could still get hardware acceleration if you needed it, with performance on par with Windows. ATI on the other hand is a choose your own adventure kind of a deal.

We’ll I did have problems, so there ?

If the GCC compiler has an issue, I'm in a lot of trouble too, even though it's open source. I guess I could scan the code and figure out any minor problems (or figure out why I was misusing it) but I'm not an expert in compilers, or even at tensor programming.

Obviously it would be better if they're open source, since someone would be able to fix them, and it helps to know how your tools really work. But a better closed source driver is probably the right tool for most problems.

It's not the case for kernel driver. Open source (and merged into main) is a big advantage in this context.

Windows and osx still do pretty well

Merged into main still sounds like a gargantuan endeavor compared to opening a support ticket

I love open source, but this is just isn't true. If you are a company, and you find an actual problem with Nvidia, and you just spent some multiple of $100k or more on graphics cards for a cluster. You are able to get an ear and get that driver fixed.

Working is moot. Their Linux support is the worst among all GPU makers.

For this particular use case, Nvidia works better on Linux than AMD

As soon as you are out of what they care about, your support will become abysmal. And they don't care about modern desktop use case (such as they didn't have Wayland support for decades).

So if you are a Linux user - simply avoid Nvidia. They don't care about you.

Nvidia's Linux compute solution is far better than AMD's compute solution, and their Linux graphics solution is far worse than AMD (and Intel)'s graphics solution. So the recommendation comes down to why you want that GPU.

Their compute solution is still lock-in, so that being better is even more moot than their general "good" Linux support.

You're looking at this from the perspective of a hobbyist, not a business who needs to get stuff done.

For workloads like training AI models, you would be shooting your business in the foot by using AMD.

> Wayland support for decades

It's only 15 years old!

And Nvidia did nothing to support until like a just a few years ago and even now their support is basically half cooked.

If it takes them that long, it's very clear they don't care about Linux in general. Same thing is reflected in them not working with upstream kernel.

Nvidia is famous for having the worst Linux drivers. Linus himself told Nvida to go fuck themselves on stage.

CUDA on Linux just works. And works very well. If anything most CUDA workloads run on Linux machines, and it's completely pain free compared to setting up ROCm even on Linux.

No, they are "famous" for having the best drivers but not open sourcing them.

NVIDIA's drivers are practically the same size as an OS anyway. Arguably they are an OS, for the GPU instead of the CPU. So it's not a huge surprise they don't give it all away for free, especially as software is a competitive advantage for them.

From what I understand that was about them not cooperating with the kernel devs. and that their driver was a black box that had to interact with half the kernel didn't help.

Driver wise my worst experiences where all solidly in the ATI/AMD binary camp, with the constant need to exorcise nouveau from various systems second.

Reply on news.ycombinator.com